The State of AI: A Guide for Information Navigators
What 2025 revealed about AI, credibility, and the work of knowing what to trust.
Every January, founding member subscribers get the Card Catalog Annual Report: a strategic assessment of AI developments from the previous year, grounded in verified data rather than press releases. This edition covers 2025, the year AI capabilities and AI marketing diverged further than ever.
Executive Summary
By 2025, the information environment had been fundamentally reshaped.
The tools people relied on to find answers increasingly summarized rather than cited, synthesized rather than showed their work. Information arrived as conclusions instead of pathways, with sources compressed or obscured in the process. For most readers, the distinction between human-written material and machine-generated text was no longer obvious in everyday use.
The scale of this change was measurable. Nearly three-quarters of newly published web pages contained AI-generated content. Roughly one in six of the top results in Google searches was written by an AI system rather than a person. As synthetic text proliferated, it blended into the background of the web, reshaping not only how information was produced, but how it was encountered and evaluated.
This transformation unfolded faster than anyone anticipated. In January 2025, a Chinese startup released an AI model that matched systems built at vastly higher cost, triggering a market selloff that erased hundreds of billions of dollars in value. By the end of the year, ChatGPT was used weekly by hundreds of millions of people. Google’s latest model topped every major benchmark. The technology improved substantially, measurably, and quickly.
And yet. Researchers at MIT found that the overwhelming majority of enterprise AI projects produced no measurable business impact. The Federal Trade Commission pursued enforcement actions against companies whose “AI-powered” products turned out to rely on human labor rather than automation. Apple demonstrated features at its developer conference that had no working implementation behind them. Across the industry, the same pattern repeated: extraordinary claims, underwhelming delivery.
The gap between what AI promises and what it delivers is fundamentally an information literacy problem. Evaluating AI claims requires the same skills that have always mattered for navigating information: assessing source credibility, verifying specific claims, understanding how systems shape what you see, calibrating trust appropriately based on evidence rather than confidence. These skills transfer directly to AI. The question is whether people recognize that they already possess the foundation for what the moment requires.
This report provides a strategic assessment grounded in verifiable evidence and primary sources, designed for readers who need to make informed decisions rather than follow hype cycles. The goal throughout is practical clarity: what happened, what it means, and how to apply these lessons in the year ahead.